AI is all the rage lately, and with good reason: it’s the most disruptive technology most of us have ever seen.
It’s rapidly changing the way we work, and while some may view it as a threat, I personally feel (just like any other disruptive technology) it’s actually an opportunity.
In this newsletter, I want to share why I feel AI is quickly becoming an essential component of a good PKM system.
The True Purpose of PKM
Personally, I believe that PKM is much more than just connecting notes.
It’s really about how information flows into and out of your world and how you make sense of that information to take consistent action on the things that really matter.
Much like a technology stack, your PKM Stack must facilitate frictionless information flow between all of the apps that you use in your PKM system.
At the top, we have the Identity level, which is your personal vision and values. The goal is to align all of the other levels of the PKM stack with this Identity level, providing the motivation to show up and take consistent action on the things that are important and the clarity to cut the things that aren’t.
But in order to do that, we need to focus on a small subset of information that is relevant to us personally.
And this requires a totally different approach than most people take to AI & LLMs.
The Problem with Large Language Models
Large Language Models (LLMs) have a fundamental problem: they’re trained on vast datasets from across the internet, mixing verified facts, unverified claims, and complete fabrications.
Because the models are trained on so much information (much of it junk), the responses are often wrong or misleading.
For example, I occasionally receive requests via email to help people troubleshoot something in Obsidian that ChatGPT has told them they can do. However, the large language model draws from information that goes far beyond the scope of the app, so it frequently creates features, plugins, and integrations that don’t really exist.
It basically assumes all information is true and useful. It really doesn’t have a filter, and frequently “hallucinates” or makes things up.
(BTW, that’s not always a bad thing. If you’re brainstorming or playing with ideas, hallucinations can be a feature, not a bug. But let’s set that aside for now 😉)
A large language model can give you answers that sound right, but it can't provide the hyper-specific, practical information that offers real value when you put it into action.
The Solution: Small Language Models
For quality answers, we need models trained on specific, curated data—essentially creating “small language models.” This mirrors how we humans operate: focusing on what matters to us while ignoring everything else.
Small language models offer significant advantages:
- Accuracy: Avoiding hallucinations by focusing on specific datasets
- Depth: Exploring niche topics in greater detail
- Relevance: Providing responses tailored to specific needs
The result? More “a-ha!” moments. More light bulbs turning on.
More illumination, less hallucination.
For example, imagine a language model trained exclusively on Obsidian for a moment. Such a model would be far more accurate when answering technical questions about how the app works when learning how to use it.
Or imagine a model trained on how to use your PKM system to do more of what matters. It would be like having 24/7 access to a private PKM coach.
Now, imagine a model that combines those two and is designed specifically to help you use Obsidian to do your best creative work.
That’s exactly what I’ve built 😉
My Custom Practical PKM Bot
I’ve created a custom AI agent inside my private community that is trained on all my course materials (Practical PKM, Obsidian 101, etc.). It can provide answers and responses that are specifically from the realm of personal knowledge management, or PKM, and gives you MUCH better responses than the LLMs can.
Here are some examples of the things you can ask:
- Can you give me a bullet-point agenda for my next Personal Retreat?
- What are some things I can do to make my new habits stick?
- Where can I download the Personal Retreat template file?
I’ve been messing with this all weekend, asking it all the questions I typically get whenever I lead a live cohort. It does an incredible job of not only answering questions but connecting with resources that will help you turn information into application (and, as a result, chaos into clarity).
But don’t just take my word for it. Here’s a quick video I recorded after trying it out for myself:
This custom AI agent is available inside the Practical PKM Hybrid Cohort, which is a combination of self-paced video lessons and live Q&A calls. I added this AI agent as a bonus, but I actually believe this might revolutionize the way people consume video courses. Being able to ask questions as you consume the material will significantly reduce the friction to take action and increase the likelihood of achieving the transformation you’re after.
In this case, creating a custom PKM system that you can actually trust to turn chaos into clarity and do your best creative work inside of Obsidian.
It's officially launching tomorrow, but the page is actually live now 🤫 You can join right now and get lifetime access for only $397 (less than half of the live cohort!) if you go to join.practicalpkm.com.
Take a Small Language Model for a Test Drive with NotebookLM
In addition to custom AI agents (which I believe will become more and more accessible), there's a great tool available right now where you can start building a small language model of your own: NotebookLM.
I've had my eye on NotebookLM for a while because of the involvement of Stephen Johnson. I'm a big fan of his book Where Good Ideas Come From, and I believe his involvement in the project means it will be more practically useful for PKM nerds like me who want to get more out of their notes and ideas.
I use the paid version myself ($20/month), but you can kick the tires with the free version, which gives you 100 notebooks and up to 50 sources (the paid version gives you up to 300 sources per notebook). But you can upload your own files and use them as source material in either one if you want to see how it works.
If you want a tool to help you understand the information that matters to you, definitely check this out.
